INTRODUCTION Ecosystems as highly complex patterns of physiogenetic, biotic, and anthropogenic factors directly or indirectly correlated with one another form a paramount functional correlation represented by 'landscape' (Leser & Rodd, 1991). The use of medium-scale (M 1: 50 000) satellite remote sensing has proven promising for the operative observation of current and historical states and for the assessment of the diversity of land cover (LC) of these landscapes. Landsat 5 TM (Thematic Mapper) sensor has many advantages in ecological applications (Cohen & Goward, 2004) because of its suitable spatial resolution (grain size associated with the grain of land management), spectral resolution (all major portions of the solar electromagnetic spectrum are represented), and temporal resolution (systematically collected remote sensing data over more than 30 years). The broad spectral range of the Landsat data offers good opportunities for the interpretation of the essential characteristics of vegetation (abundance, state of biomass, etc.), subsoil character, and, importantly, the water content of both. This is especially important in territories covered with peatlands, where the percentage of moisture is high in the vegetation and moss surface. Moreover, information from Landsat satellites has by far the best cost benefit ratio. In land use/land cover satellite monitoring studies high-resolution satellite images (IKONOS, QuickBird, SPOT5) offer a much greater potential for accurate vegetation mapping (Ozdemir et al., 2005), especially in mires (Langanke et al., 2007). However, they are more expensive than Landsat and not available from earlier years. Identification of meaningful biogeophysical features in satellite images requires the existence of a consistent and universal land cover nomenclature (LCN) for the observed patterns. The world-wide land use and land cover classification scheme (Anderson et al., 1976) needs to be adapted to local environmental conditions and should encompass all natural, semi-natural, and man-made patterns that can be recognized from satellite images. Their dynamics, followed on multi-date images, will reflect the environmental development. In 1996 a national project entitled 'Remote Sensing of Estonian Landscapes' (RSEL), focusing on the monitoring of landscapes in selected nature protection areas from medium-scale (M 1: 50 000) satellite images, was launched. Monitoring paid special attention to protected areas, but the surroundings were included to provide for 'neighbourhood ecology' (Forman, 1995; Kintz et al., 2006). The monitoring sites in Estonia consist therefore of protected (core) and reconciled 3 km wide buffer zones around them to better satisfy the needs of nature management. The objectives of the present work were: 1. Development of a methodology for highly selective LC recognition from satellite images by using GIS technologies and field work 2. Computer-aided classification of LC patterns in satellite images using classification masks 3. Highlighting qualitative and quantitative changes in the monitoring sites 4. Predicting LC development trends in monitoring sites 5. Selecting diversity metrics for the studying of landscape diversity. This work is an extension of similar studies by using aerial photos (Aaviksoo, 1988, 1993a) and monitoring and modelling LC dynamics by using Markov Models (MM) (Aaviksoo, 1995a). Recently, several similar studies have mapped LC and observed changes in protected areas, nature reserves, or natural parks (Poulin et al., 2002; Groom et al., 2005; Hilbert, 2006; Von Wehrden et al., 2006) and predicted their future trends using the Markovian approach (Flamenco-Sandoval et al., 2007) to name but a few. The present paper puts emphasis on a comprehensive approach, meaning the use of all available national and local GIS-based data, allowing us to archive a highly selective satellite mapping of the protected areas. …
Read full abstract